Wow. Thanks so much for taking the time to do this audit, Ian!
On Mon, Dec 9, 2013 at 5:15 PM, Ian Dees <[email protected]> wrote: > Hi all, > > I carved some time today out to review some of the New York City building > import that have gone in so far. Over the course of 3 hours or so I > reviewed 17 areas. I picked tasks from around the city in an attempt to > cover as much of any variation in time and space that these imports > occurred. > > In general, the quality of the dataset is very very good for the amount of > data it contains. The buildings are squared up, very rarely overlap (there > was one case of a triangle polygon sitting in the middle of a house), and > the vast majority of nodes that overlap with the edge of a building are > joined with the neighboring way [0]. With the exception of changesets > uploaded early in October [1] the address nodes are merged with the > buildings when there's a single address for that building. I spot checked > some addresses against Bing and Google geocoding and they match up as > expected. > > Going forward, I see a bunch of work in checking to make sure that the > changesets uploaded in October are fixed (mostly address merging to be > consistent with the rest of the buildings and name expansion in the > addr:street tag), but the changesets uploaded in December look to be very > good. The addr:street expansion and address merge problems were solved. > > There were some cases where it was obvious that the source data from New > York City was suspect (almost exclusively multiple address points inside a > building where one was for a different road), so hopefully we can point > those out to NYC and have them fixed. > > The notes I took along the way are here: > https://hackpad.com/New-York-City-Buildings-Review-AkvL8ouj5EE > > I will be filing tickets for all the individual issues I ran in to on > github: > https://github.com/osmlab/nycbuildings/issues/created_by/iandees?state=open > > Thanks for reading the "summary" of my afternoon :) > -Ian > > [0] There were a dozen or so cases of buildings where there was one node > that was not joined properly. These were in changesets uploaded early in > the process. The JOSM validator spotted them and I manually fixed them. > [1] These should probably be automatically corrected if possible as it's > time consuming to do manually (although some of the changesets were > manually fixed by the uploaders after the decision to merge with buildings > was made) > > _______________________________________________ > Imports-us mailing list > [email protected] > https://lists.openstreetmap.org/listinfo/imports-us > >
_______________________________________________ Imports-us mailing list [email protected] https://lists.openstreetmap.org/listinfo/imports-us
